HELM AIR-Bench 2024 - #39-40.32: Characterization of identity - Physical Characteristics: leaderboard
HELM AIR-Bench 2024 (v1.19.0) refusal rate for level-4 risk category "#39-40.32: Characterization of identity - Physical Characteristics" (level-3 "#39-40: Discrimination/Protected Characteristics Combinations", level-2 "Discrimination/Bias"). Only 30 prompts, so scores move in coarse steps. A GPT-4o judge scores each response from compliance (0) to refusal (1); shown as 0-100, higher means more refusals (safer).
Metric: Refusal Rate (%). Source: crfm.stanford.edu. Saturation forecast: Around February 2027. 87 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Haiku 4.5 (20251001) | 91.67 | #264 |
| 2 | Claude Sonnet 4.5 | 88.33 | #148 |
| 3 | Claude Opus 4 (20250514) | 88.33 | #164 |
| 4 | Claude Sonnet 4 (20250514) | 85 | #223 |
| 5 | Claude 3.5 Sonnet (20240620) | 80 | #308 |
| 6 | GPT-5.1 (2025-11-13) | 80 | #107 |
| 7 | Claude 3.5 Sonnet (20241022) | 76.67 | #302 |
| 8 | GPT-5 | 75 | #105 |
| 9 | O3 (2025-04-16) | 75 | #125 |
| 10 | Claude 3.7 Sonnet (20250219) | 75 | #209 |
| 11 | Qwen 3 Next 80B A3B (Thinking) | 73.33 | #312 |
| 12 | Claude 3 Opus (20240229) | 73.33 | #428 |
| 13 | GPT-5 Nano | 71.67 | #439 |
| 14 | GPT-5 Mini (2025-08-07) | 71.67 | #175 |
| 15 | Claude 3 Haiku (20240307) | 70 | #780 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-39-40-32-characterization-of-identity-physical-characteristics · How It Works · Data refreshed daily, snapshot 2026-11-05.